Research On Image Segmentation Based On CV Level Set Method

In the field of image segmentation,the active contour image segmentation models based on level set method have made great achievement.This kind of method which focuses on the image region information has advantages for lower computation,smooth topology,multi-information communicability.In recent years,the level set method image segmentation models have received attention by the computer vision scholars at home and abroad.The energy function is applied to map the curve of the low dimension space to the high space as well as the curve is evolved by the energy function in the level set method.The level set method utilizes the partial differential equations to evolve the contour instead of computing the contour parameter equation in order to reduce the computational complexity.The image segmentation models based on the level set method fall into two major categories:the models based on the global region information and the models based on the local region information.The global region-based models such as Chan-Vese model?hereinafter called by CV model?have advantage for noise resisting and lower computation complexity,however,the CV model and the related models fail to segment the images with intensity inhomogeneity.The local region-based models such as Region-scalable Fitting model?hereinafter called by RSF model?overcome the intensity inhomogeneity,nevertheless,the local region-based models are sensitive to the initial contour,in other words,the uncomfortable initial contour may cause the segmentation to fail.This paper mainly does study the above-mentioned problems.In order to overcome the image segmentation difficulty caused by the initial contour,the novel image segmentation model introduce the clustering algorithm to divide the image pixels as well as the position of the initial contour is able to be obtained.The target region image gray-mean value and the background region image gray-mean value can be obtained to construct novel energy terms in order to introduce the image gray-mean value of the initial region to the novel energy function which improves the traditional image segmentation model.The novel energy function combines Chan-Vese model and Region-scalable Fitting model to reduce the noise influence,as well as the proportion of each novel energy term is able to be adjusted by the parameters.The main research context is as follows:?1?GMMCV model based on the Gaussian mixture model and CV level set method is proposed in this paper.GMMCV model introduces the Gaussian mixture model to divide the image pixels for the sake of solving the problem caused by the initial contour.The image region can be divided into the target region and the background region,moreover,the edge between the two regions can be defined the initial contour.In order to introduce the gray mean of the regions to GMMCV model,the gray-mean value of statics in these two regions need to be obtained.The gray-mean value of statics can be updated with the curve evolving.Finally,the contour will evolve to truth target region.In order to test the proposed model and compare with other related models,the experimental data set employed BSDS500 natural scene images.The experimental results show that GMMCV model is able to overcome the initial contour sensitive problem as well as the proposed model improve the accuracy of the segmentation.?2?The adaptive RSF-CV model based on K-Means clustering algorithm,RSF and CV model is proposed in this paper.Introducing K-Means clustering algorithm which has advantage for reducing the image noise influence and the intensity inhomogeneity,to the adaptive RSF-CV model for dividing the image into target region and background region.The initial contour is able to be obtained by the boundary between the target region and the background region.The RSF-CV model employs the means of the K-Means clusters as the gray-mean values of the target region and the background region.In order to solve problem that image intensity inhomogeneity,a novel energy function consist of Chan-Vese model and Region-scalable model which are given the parameters.The parameters that represent the proportion of the CV model and the RSF model in the energy function can be adjusted through the algorithm related to LBP method.?={?₁,?₂,?₃...} is represented as the difference between the intensity of pixel and the intensity of other pixels around it,at the same time,?? can be obtained through the gray-mean values of the target region and the background region.The number of themvalue which is greater than the ?? value can be figured out,otherwise,it can be utilized to adjust the parameters so as to make the energy terms play role in the curve evolving.The experiment,that shows the adaptive RSF-CV model and other related image segmentation model experimental result,has been performed on natural scene data set BSDS500 and magnetic resonance angiography.The experimental results show that the method proposed in this paper is able to deal with noisy and intensity inhomogeneity images effectively.